A A MERICAN MERICAN Washington, Suite 1430 (ISSN American K 600 0003-1224) Street

S S DC OCIOLOGICAL OCIOLOGICAL NW Sociological 20005 Review Scholars from around the world will travel to the ASA Annual Meeting in San Francisco this R R EVIEW EVIEW summer. The 104th Annual Meeting will be an intellectual conversation that offers focused educational sessions by leading scholars, networking opportunities, and professional develop- ment solutions from experts in the field. Connect, learn, and share with colleagues as we explore the program theme: The New Politics of Community.

Engage Your Mind... VOLUMEVOLUME 7474 •• NUMBERNUMBER 33 •• JUNEJUNE 20092009 OFFICIALOFFICIAL JOURNALJOURNAL OFOF THETHE AMERICANAMERICAN SOCIOLOGICALSOCIOLOGICAL ASSOCIATIONASSOCIATION Opening Plenary Session. How Communities Matter: Perspectives of Artists, Academics and Activists IINEQUALITYNEQUALITY ANDAND SSURVEILLANCEURVEILLANCE (Friday, August 7, 7:00-9:00pm) AliceAlice GoffmanGoffman Plenary Session. Why Obama Won (and What That Says about Democracy WantedWanted MenMen inin aa PhiladelphiaPhiladelphia GhettoGhetto and Change in America) J J

(Saturday, August 8, 12:30-2:10pm) UNE UNE TTHEORYHEORY ANDAND MMODELINGODELING

ASA Awards Ceremony and Presidential Address by Patricia Hill Collins 2009 2009 NeilNeil GrossGross (Sunday, August 9, 4:30-6:10pm) AA PragmatistPragmatist TheoryTheory ofof SocialSocial MechanismsMechanisms

Plenary Session. Bringing Communities Back In: Setting a New Policy CristobalCristobal YoungYoung Agenda ModelModel UncertaintyUncertainty inin SociologicalSociological ResearchResearch (Monday, August 10, 12:30-2:10pm) WWELFAREELFARE ASA President Patricia Hill Collins and the Program Committee have also organized a mini- symposium, a meeting within the general meeting, which explores how the election of Barack SanfordSanford F.F. Schram,Schram, JoeJoe Soss,Soss, RichardRichard C.C. Fording,Fording, andand LindaLinda HouserHouser Obama might signal a new politics of community in action. This mini-symposium consists of a Race,Race, Choice,Choice, andand PunishmentPunishment atat thethe FrontlinesFrontlines ofof WelfareWelfare ReformReform cluster of sessions that are scheduled throughout the meetings featuring noted scholars such as Melissa Harris Lacewell, professor of political science at Princeton University; Gurminder K. Bambra of the University of Warwick; and Peter Levine, director of CIRCLE (the Center for JenniferJennifer VanVan HookHook andand FrankFrank D.D. BeanBean Information Research on Civic Learning and Engagement). CulturalCultural RepertoiresRepertoires andand WelfareWelfare BehaviorsBehaviors V V The Annual Meting program also features more than 600 sessions representing some of the OL OL DDELINQUENCYELINQUENCY best of emerging and cutting-edge research. . . 74, 74, DavidDavid J.J. HardingHarding ...Represent Your Discipline...Be a Part of the Action N N Violence,Violence, AdolescentAdolescent Boys,Boys, andand OlderOlder PeersPeers O O . . 3, 3,

For more information, visit: www.asanet.org. Register and Book Your Hotel Today! additional

Periodicals RichardRichard J.J. PettsPetts at P P Washington, Family,Family, Religion,Religion, andand DelinquencyDelinquency TrajectoriesTrajectories P P . . 339–505 339–505 mailing

postage DavidDavid S.S. KirkKirk DC

offices ResidentialResidential ChangeChange andand Recidivism:Recidivism: LessonsLessons fromfrom HurricaneHurricane KatrinaKatrina paid and

Electronic copy available at: http://ssrn.com/abstract=1413908 ASR EDITORS VINCENT J. ROSCIGNO RANDY HODSON Ohio State Ohio State DEPUTY EDITORS DOUGLAS DOWNEY PAMELA PAXTON Ohio State Ohio State FRANCE WINDDANCE TWINE It’s Here... STEVEN MESSNER ZHENCHAO QIAN UC, Santa Barbara SUNY-Albany Ohio State GEORGE WILSON LINDA D. MOLM VERTA TAYLOR U. of Miami Arizona UC, Santa Barbara The 2009 Guide EDITORIAL BOARD PATRICIA A. ADLER RACHEL L. EINWOHNER CECILIA MENJIVAR ROGELIO SAENZ to Graduate U. of Colorado Purdue Arizona State Texas A&M OSEPH ALASKIEWICZ ICHAEL ESSNER SIGAL ALON J G M A. M EVA N SCHOFER Arizona USC Tel-Aviv U. UC, Irvine Departments PAUL E. BELLAIR JENNIFER GLICK DAVID S. MEYER Arizona State UC, Irvine VICKI SMITH Ohio State PEGGY C. GIORDANO JOYA MISRA UC, Davis KENNETH A. BOLLEN Bowling Green Massachusetts SARAH SOULE of Sociology UNC, Chapel Hill KAREN HEIMER JAMES MOODY Stanford CLEM BROOKS Iowa Duke Indiana MAXINE S. THOMPSON This invaluable reference has been DENNIS P. HOGAN CALVIN MORRILL North Carolina State CLAUDIA BUCHMANN Brown UC, Irvine published by the ASA annually since Ohio State MATT L. HUFFMAN ORLANDO PATTERSON STEWART TOLNAY 1965. A best seller for the ASA for JOHN SIBLEY BUTLER UC, Irvine Harvard Washington many years, the Guide provides com- UT, Austin GUILLERMINA JASSO TROND PETERSEN HENRY A. WALKER PRUDENCE CARTER NYU UC, Berkeley Arizona prehensive information for academic Harvard GA RY F. JENSEN ALEJANDRO PORTES MARK WARR administrators, advisors, faculty, stu- Vanderbilt Princeton WILLIAM C. COCKERHAM UT, Austin dents, and a host of others seeking Alabama–Birmingham ANNETTE LAREAU JEN’NAN G. READ Maryland Duke BRUCE WESTERN information on social science depart- MARK COONEY EDWARD J. LAWLER LINDA RENZULLI Harvard Georgia ments in the United States, Canada, Cornell Georgia MARK WESTERN WILLIAM F. DANAHER and abroad. Included are listings for JENNIFER LEE BARBARA JANE RISMAN U. of Queensland College of Charleston UC, Irvine UIC 224 graduate departments of sociol- ZHENG WU JAAP DRONKERS J. SCOTT LONG WILLIAM G. ROY European U. Institute Indiana UC, Los Angeles University of Victoria ogy. In addition to name and rank, MITCHELL DUNEIER HOLLY J. MCCAMMON DEIRDRE ROYSTER JONATHAN ZEITLIN faculty are identified by highest de- Princeton and CUNY Vanderbilt William and Mary Wisconsin gree held, institution and date of de- gree, and areas of specialty interest. MANAGING EDITOR Special programs, tuition costs, types of financial aid, and student enroll- MARA NELSON GRYNAVISKI ment statistics are given for each department, along with a listing of recent EDITORIAL COORDINATOR EDITORIAL ASSOCIATES PhDs with dissertation titles. Indices of faculty, special programs, and PhDs STEPHANIE A. WAITE ANNE E. MCDANIEL SALVATORE J. RESTIFO LINDSEY PETERSON CAROLYN SMITH KELLER awarded are provided. 424 pages. JONATHAN VESPA ASA Order online at www.asanet.org/bookstore. MEMBERS OF THE COUNCIL, 2009 Member Price: $ 30.00 PATRICIA HILL COLLINS, MARGARET ANDERSEN, DONALD TOMASKOVIC-DEVEY, Non-member Price: $ 50.00 PRESIDENT VICE PRESIDENT SECRETARY Student Member Price: $ 20.00 Maryland Delaware UMass, Amherst

EVELYN NAKANO GLENN, JOHN LOGAN, ARNE KALLEBERG, DOUGLAS MCADAM, American Sociological Association PRESIDENT-ELECT VICE PRESIDENT-ELECT PAST PRESIDENT PAST VICE PRESIDENT 1430 K Street NW, Suite 600 UC, Berkeley Brown UNC Stanford Washington, DC 20005 SALLY T. HILLSMAN, Executive Officer (202) 383-9005 • Fax (202) 638-0882 [email protected] • www.asanet.org ELECTED-AT-LARGE

DALTON CONLEY PIERRETTE HONDAGNEU-SOTELO MA RY PATTILLO RUBEN RUMBAUT NYU USC Northwestern UC, Irvine MARJORIE DEVAULT OMAR M. MCROBERTS CLARA RODRIGUEZ MARC SCHNEIBERG Syracuse U. of Chicago Fordham Reed ROSANNA HERTZ DEBRA MINKOFF MA RY ROMERO ROBIN STRYKER Wellesley Barnard Arizona State Minnesota

Electronic copy available at: http://ssrn.com/abstract=1413908 ARTICLES 339 On the Run: Wanted Men in a Philadelphia Ghetto Alice Goffman

358 A Pragmatist Theory of Social Mechanisms Neil Gross

380 Model Uncertainty in Sociological Research: An Application to Religion and Economic Growth Cristobal Young

398 Deciding to Discipline: Race, Choice, and Punishment at the Frontlines of Welfare Reform Sanford F. Schram, Joe Soss, Richard C. Fording, and Linda Houser

423 Explaining Mexican-Immigrant Welfare Behaviors: The Importance of Employment-Related Cultural Repertoires Jennifer Van Hook and Frank D. Bean

445 Violence, Older Peers, and the Socialization of Adolescent Boys in Disadvantaged Neighborhoods David J. Harding

465 Family and Religious Characteristics’ Influence on Delinquency Trajectories from Adolescence to Young Adulthood Richard J. Petts

484 A Natural Experiment on Residential Change and Recidivism: Lessons from Hurricane Katrina David S. Kirk

V OLUME 74 • NUMBER 3 • JUNE 2009 O FFICIAL J OURNAL OF THE A MERICAN S OCIOLOGICAL A SSOCIATION The American Sociological Review (ISSN 0003-1224) is published bimonthly in February, April, June, August, October, and December by the American Sociological Association, 1430 K Street NW, Suite 600, Washington, DC 20005. Phone (202) 383- 9005. ASR is typeset by Marczak Business Services, Inc., Albany, New York and is printed by Edwards Brothers, Inc., Lillington, North Carolina. Periodicals postage paid at Washington, DC and additional mailing offices. Postmaster: Send address changes to the American Sociological Review, 1430 K Street NW, Suite 600, Washington, DC 20005.

SCOPE AND MISSION The American Sociological Review publishes original (not previously published) works of interest to the discipline in gen- eral, new theoretical developments, results of qualitative or quantitative research that advance our understanding of funda- mental social processes, and important methodological innovations. All areas of sociology are welcome. Emphasis is on exceptional quality and general interest.

MANUSCRIPT SUBMISSION ETHICS: Submission of a manuscript to another professional journal while it is under review by the ASR is regarded by the ASA as unethical. Significant findings or contributions that have already appeared (or will appear) elsewhere must be clear- ly identified. All persons who publish in ASA journals are required to abide by ASA guidelines and ethics codes regarding plagiarism and other ethical issues. This requirement includes adhering to ASA’s stated policy on data-sharing: “Sociologists make their data available after completion of the project or its major publications, except where proprietary agreements with employers, contractors, or clients preclude such accessibility or when it is impossible to share data and protect the confi- dentiality of the data or the anonymity of research participants (e.g., raw field notes or detailed information from ethnographic interviews)” (ASA Code of Ethics, 1997).

MANUSCRIPT FORMAT: Manuscripts should meet the format guidelines specified in the Notice to Contributors published in the February and August issues of each volume. The electronic files should be composed in MS Word, WordPerfect, or Excel. All text must be printed double-spaced on 8-1/2 by 11 inch white paper. Use Times New Roman, 12-point size font. Margins must be at least 1 inch on all four sides. On the title page, note the manuscript’s total word count (include all text, references, and footnotes; do not include word counts for tables or figures). You may cite your own work, but do not use wording that identifies you as the author.

SUBMISSION REQUIREMENTS: Submit two (2) print copies of your manuscript and an abstract of 150 to 200 words. Include all electronic files on floppy disk or CD. Enclose a $25.00 manuscript processing fee in the form of a check or money order payable to the American Sociological Association. Provide an e-mail address and ASR will acknowledge the receipt of your manuscript. In your cover letter, you may recommend specific reviewers (or identify individuals ASR should not use). Do not recommend colleagues, collaborators, or friends. ASR may choose to disregard your recommendation. Manuscripts are not returned after review.

ADDRESS FOR MANUSCRIPT SUBMISSION: American Sociological Review, The Ohio State University, Department of Sociology, 238 Townshend Hall, 1885 Neil Avenue Mall, Columbus, OH 43210-1222 ([email protected], 614-292-9972).

EDITORIAL DECISIONS: Median time between submission and decision is approximately 12 weeks. Please see the ASR journal Web site for more information (http://www2.asanet.org/journals/asr/).

ADVERTISING, SUBSCRIPTIONS, AND ADDRESS CHANGES

ADVERTISEMENTS: Submit to Publications Department, American Sociological Association, 1430 K Street NW, Suite 600, Washington, DC 20005; (202) 383-9005 ext. 303; e-mail [email protected].

SUBSCRIPTION RATES: ASA members, $40; ASA student members, $25; institutions, $196. Add $20 for postage outside the U.S./Canada. To subscribe to ASR or to request single issues, contact the ASA Subscriptions Department ([email protected]).

ADDRESS CHANGE: Subscribers must notify the ASA Executive Office ([email protected]) six weeks in advance of an address change. Include both old and new addresses. Claims for undelivered copies must be made within the month fol- lowing the regular month of publication. When the reserve stock permits, ASA will replace copies of ASR that are lost because of an address change.

Copyright © 2009, American Sociological Association. Copying beyond fair use: Copies of articles in this journal may be made for teaching and research purposes free of charge and without securing permission, as permitted by Sections 107 and 108 of the United States Copyright Law. For all other purposes, permission must be obtained from the ASA Executive Office. (Articles in the American Sociological Review are indexed in the Abstracts for Social Workers, Ayer’s Guide, Current Index to Journals in Education, International Political Science Abstracts, Psychological Abstracts, Social Sciences Index Sociological Abstracts, SRM Database of Social Research Methodology, United States Political Science Documents, and University Microfilms.)

The American Sociological Association acknowledges with appreciation the facilities and assistance provided by The Ohio State University. ASR EDITORIAL TRANSITION The Council of the American Sociological Association has announced the selection of Tony N. Brown, Katharine M. Donato, Larry W. Isaac, and Holly J. McCammon as Editors-Elect of the American Sociological Review. On January 1, 2010, the term of the current editors ends and Professors Brown, Donato, Isaac, and McCammon become ASR editors. The following is the address for the new ASR editorial office:

American Sociological Review PMB 351803 2301 Vanderbilt Place Nashville, TN 37235 E-mail: [email protected] Phone: (615) 343-0426

When submitting manuscripts for review, authors should observe the following graduated transition schedule:

NEW MANUSCRIPTS As of July 1, 2009, send all new manuscripts to Editors-Elect Brown, Donato, Isaac, and McCammon. They will make editorial decisions on all new manuscripts submitted to them.

REVISED MANUSCRIPTS Send all revised manuscripts to Editors Roscigno and Hodson at the current Ohio State University address until August 7.

AFTER AUGUST 7, 2009: Send all new and revised manuscripts (even revisions invited by Professors Roscigno and Hodson) to Editors-Elect Brown, Donato, Isaac, and McCammon.

NEW SUBMISSION PROCESS: The ASR will be moving to electronic submissions sometime after July 1, 2009. Starting July 1, please check the ASR Web site (http://www2.asanet.org/journals/asr/) for further updates and instructions before submitting your manuscript. Deciding to Discipline: Race, Choice, and Punishment at the Frontlines of Welfare Reform

Sanford F. Schram Joe Soss Bryn Mawr College University of Minnesota

Richard C. Fording Linda Houser University of Kentucky Bryn Mawr College

Welfare sanctions are financial penalties applied to individuals who fail to comply with welfare program rules. Their widespread use reflects a turn toward disciplinary approaches to poverty management. In this article, we investigate how implicit racial biases and discrediting social markers interact to shape officials’decisions to impose sanctions. We present experimental evidence based on hypothetical vignettes that case managers are more likely to recommend sanctions for Latina and black clients—but not white clients—when discrediting markers are present. We triangulate these findings with analyses of state administrative data. Our results for Latinas are mixed, but we find consistent evidence that the probability of a sanction rises significantly when a discrediting marker (i.e., a prior sanction for noncompliance) is attached to a black rather than a white welfare client. Overall, our study clarifies how racial minorities, especially African Americans, are more likely to be punished for deviant behavior in the new world of disciplinary welfare provision.

isciplinary approaches to poverty man- tice appropriate behaviors (Schram 2006). Dagement are ascendant in the United States Today, public aid programs are more directive today. They are perhaps most visible in the area in setting behavioral expectations, more super- of criminal justice, where tough new policies visory in monitoring compliance, and more have driven incarceration rates to levels that punitive in responding to infractions (Mead are unprecedented in U.S. history and unrivaled 1997). “New paternalist” welfare programs use by other nations (Western 2006). Yet mass incar- a variety of incentives, surveillance mecha- ceration is far from an isolated development nisms, and restrictive rules to modify client (Starobin 1998). Welfare policies for the poor behaviors. A system for dispensing punishments have been redesigned in recent years to reflect when incentives and rules prove insufficient the idea that the state has a legitimate interest serves as a failsafe in this process. Sanctions— in ensuring that socially marginal groups prac- penalties that reduce or terminate benefits in

Direct all correspondence to Sanford F. Schram, iment; Norma Altshuler, Jeremy Babener, and Helen Graduate School of Social Work and Social Research, Tang for commenting on the vignettes we used; and Bryn Mawr College, 300 Airdale Road, Bryn Mawr, Vicki Lens, Suzanne Mettler, and anonymous review- PA 19010-1697 ([email protected]). The ers for extensive and insightful commentary on ear- authors thank Tony Chen, Sandy Danziger, Sheldon lier versions of this analysis. Thanks also to staff at Danziger, Phoebe Ellsworth, Mary Jo Deegan, Ann Florida’s Agency for Workforce Innovation and the Lin, Vince Hutchings, Julia McQuillan, Mark Peffley, Department of Children and Families. The research Leslie Rescorla, Marc Schulz, Kristen Seefeldt, Anjali reported here was supported by the Annie E. Casey Thapar, Loic Wacquant, and Kathy Cramer Walsh for Foundation (grant # 205.0059), the University of commenting on our research; David Consiglio for Kentucky Center on Poverty Research, the Institute helping design and administer our Web-based exper- for Research on Poverty, and Bryn Mawr College.

AMERICAN SOCIOLOGICAL REVIEW, 2009, VOL. 74 (June:398–422) DECIDING TO DISCIPLINE 399 response to client noncompliance—are the ulti- sanctioning, where the actions of frontline work- mate expression of welfare’s disciplinary turn ers are bracketed in favor of attention to client and deserve the greatest credit for the exodus of and community characteristics (Pavetti, Derr, clients from welfare since the mid-1990s and Hesketh 2003; Wu et al. 2006). Our study (Rector and Youssef 1999). bridges these two streams by asking how client This article investigates welfare administra- race influences caseworker decisions to apply tors’ decisions to sanction clients, focusing on penalties. We break new ground, first, by pre- how the interplay of racial status and character senting rigorous empirical tests of racial bias in markers shapes such decisions. Our analysis welfare sanction decisions and, second, by pre- falls into a long tradition of sociological research senting an explicit cognitive model that explains on the ways that major societal institutions pro- how client race exerts an influence on officials’ duce racial disparities. Using observational and decisions to punish. audit designs, researchers in recent years have To theorize the influence of race, we draw on revealed stark disparities in the ways black and the Racial Classification Model (RCM) of pol- white Americans are treated in retail transactions icy choice, which we developed and tested on (Ayres and Siegelman 1995), the mortgage loan state-level policy choices elsewhere (Soss, industry (Munnell et al. 1996), the insurance Fording, and Schram 2008). Drawing on mod- industry (Wissoker, Zimmermann, and Galster els of implicit racism (Quillian 2008), the RCM 1998), the healthcare sector (Schulman et al. specifies conditions that can produce policy- 1999), housing markets (Yinger 1995), and based racial disparities even in the absence of labor markets (Bertrand and Mullainathan 2003; racial animus or discriminatory intent. In this Pager 2007). Despite the broad scope of this article, we test hypotheses derived from the research, however, there are few systematic RCM’s assertion that racial disparities become studies of how race matters when government more likely when policy targets possess dis- agents use their distinctive authority to punish. crediting traits consistent with negative group The major exception, of course, is the literature reputations. Extending the work of Pager (2007), exploring the origins and consequences of racial we show how stereotype-consistent markers disparities in arrest, sentencing, and incarcera- can provide “expectancy confirmation” (Darley tion patterns (Pager 2007; Western 2006). It is and Gross 1983) and, hence, strengthen the link unclear, however, how findings in the criminal between racial status and policy treatment justice domain might generalize to the welfare (Correll, Benard, and Paik 2007). As Duster domain or to other policy areas. (2008) notes, much of the evidence for implic- We shift the analytic focus on punishment, in it racism comes from artificial research set- Bourdieu’s (1998) terms, from the “right hand tings; there is a pressing need for empirical of the state” to the “left hand of the state”—or tests in real-world settings and for models that more precisely, from criminal justice to social explain how implicit cognitive biases translate welfare provision. It is well established that into concrete policy outcomes. racial attitudes influence public preferences To meet these goals, we test propositions regarding welfare policy (Gilens 1999) and that from the RCM using both experimental and state-level welfare policy choices closely track administrative data. Our experimental data come the racial composition of welfare caseloads from a Web-based survey of Florida Welfare (Soss et al. 2001). But welfare case managers Transition (WT) case managers. Respondents work directly with clients in concrete organi- were presented with realistic rule-violation zational settings; as a population, they are quite vignettes in which key client characteristics unlike the “average” citizen or state legislator. were randomly assigned. Case managers were Their decisions to punish specific individuals then asked whether they would impose a sanc- and families are a far cry from expressions of tion in response. To our knowledge, no prior general policy preferences or choices among sanctioning study has employed this approach, state policy alternatives. which offers the advantages of causal inference Although there is a significant ethnographic associated with experiments (Kinder and Palfrey literature on race and welfare case managers 1993) while remaining close to the phenomenon (Bonds 2006; Watkins-Hayes 2009), its insights of study: decisions made by actual case man- have had little impact on econometric studies of agers. 400 AMERICAN SOCIOLOGICAL REVIEW

Despite the advantages of this survey-exper- There is extensive evidence that sanctions imental design, hypothetical vignettes and the have played a key role in transforming welfare limitations of our sample both counsel some from a system focused on cash benefits to one humility regarding these data. Accordingly, we tri- focused on enforcing work (see Pavetti et al. angulate our findings with administrative data 2004). Between 1997 and 1999, nearly 500,000 from the Florida WT program. While these data families lost benefits due to sanctions—approx- offer a weaker basis for causal inference, they imately one quarter of the caseload reduction for have the benefit of reflecting decisions made on that period (Goldberg and Schott 2000). Indeed, actual cases under normal working conditions. the states with the strictest sanctions experi- enced caseload declines as much as 25 percent greater than those reported by states with the SANCTIONS AND WELFARE REFORM least stringent sanctions (Rector and Youssef The use of sanctions to enforce client compli- 1999). Some suggest that the threat of sanc- ance predates welfare reform. Prior to 1996, tions may be responsible for even greater num- however, sanctions were used infrequently and bers of recipients leaving the rolls (Lindhorst, were applied only to the benefits of the adult Mancoske, and Kemp 2000). Moreover, studies head of a household, not the entire family suggest that being sanctioned significantly (Bloom and Winstead 2002). Sanctions became increases hardship among recipients (Reichman, far more important under the Temporary Teitler, and Curtis 2005; Stahl 2008). Assistance for Needy Families (TANF) program Together, these studies suggest the vital because national welfare reform legislation in importance of understanding how sanction deci- 1996 specified stricter work requirements, set sions actually get made at the frontlines of wel- narrower exemption criteria, made a broader fare reform. Direct responsibility for these scope of behaviors subject to sanction, and gave decisions falls to individual caseworkers who states more options in designing penalties must respond to individual cases. Such workers (Hasenfeld, Ghose, and Larson 2004). Perhaps have long held discretion in dealing with their most important, state TANF programs now have clients (Lipsky 1980; Maynard-Moody and to meet specific quotas for the percentage of Musheno 2003), but welfare reform has given recipients participating in work-related activi- them a variety of new powers and responsibil- ties. Sanctions are the primary disciplinary tool ities (Watkins-Hayes 2009). To understand how available to administrators as they seek to moti- race matters for punishment in the new world of welfare, one must theorize and investigate vate client behaviors to meet these quotas. case-manager decision making. Sanctions are most often applied when recipi- ents fail to complete required hours of partici- pation in countable work activities such as job THE RACIAL CLASSIFICATION search, job-readiness classes, vocational edu- MODEL AND DECISIONS TO PUNISH cation, training, community work, and paid Our subject stands in the shadow of a long his- employment. torical relationship between race and welfare State sanction policies vary considerably provision in the United States (Lieberman 1998; (Bloom and Winstead 2002). To capture this Quadagno 1994). Many studies cast doubt on variation, Pavetti and colleagues (2003) offer a the idea that contemporary welfare reform rep- simplifying typology. Seventeen states rely on resents a break with this troubled history the strictest combination of choices, enforcing (Neubeck and Cazenave 2001; Schram, Soss, an “immediate full family sanction.” In these and Fording 2003); race-coded appeals and states, the entire TANF family is removed from racialized public responses played a key role in the rolls at the first instance of noncompliance. the national debates leading up to reform Eighteen states use a “gradual full-family sanc- (Hancock 2004; Reese 2005). Under devolution, tion,” which potentially has the same effect, but the racial composition of welfare caseloads has only after continued noncompliance. The been a strong predictor of state choices regard- remaining states enforce a “partial sanction” ing welfare rules and governance arrangements of benefits (usually reducing only the adult por- (Soss et al. 2001; Soss et al. 2008). tion of the grant). Implementation studies suggest that client race DECIDING TO DISCIPLINE 401 may affect caseworker decisions in sanctioning become more salient in a policy context as racial and other areas (Goldrick-Rab and Shaw 2005; minorities figure more prominently in policy- Gooden 2003; Kalil, Seefeldt, and Wang 2002). relevant political events, media discourses, and But how and why should client racial char- target-group images. Under such conditions, acteristics influence decisions to sanction wel- the social classifications that guide policy fare recipients? To answer this question, we actions—whether aimed at groups or individu- draw on the RCM. The RCM offers an expla- als—are more likely to be based on racial cat- nation for how race and ethnicity influence indi- egories. Welfare policy offers a case in point viduals’policy choices in an era in which de jure (Gilens 1999; Schram et al. 2003). discrimination is outlawed and egalitarian norms 3. The likelihood of racially patterned poli- are widely endorsed. Drawing on theories of cy outcomes will be positively associated with implicit racism (Quillian 2008), the RCM the degree of policy-relevant contrast in policy asserts that racial disparities in policy domains actors’perceptions of racial groups. The degree can emerge from cognitive biases in decision of contrast, in turn, will be a function of (1) the making even in the absence of conscious racial prevailing cultural stereotypes of racial groups, animus, out-group threat, or in-group favoritism (2) the extent to which policy actors hold rele- (cf. Blalock 1967; Key 1949). vant group stereotypes, and (3) the presence or An extension of the theory of target popula- absence of stereotype-consistent cues. tions (Schneider and Ingram 1997), the RCM It is the contrast between group reputations that focuses on how social classifications and group allows racial markers to underwrite broad reputations guide decisions about how to design assumptions about individual clients and target and implement policy tools. The model con- groups. The third premise of the RCM suggests sists of three basic propositions, which we that such contrasts depend on differences in racial explain in detail elsewhere (Soss et al. 2008). We groups’cultural images, differences in how pol- briefly review them here to clarify the basis of icy actors internalize these images, and proximate our hypotheses. cues that can evoke racial stereotypes and sug- 1. To be effective in designing policies and gest whether a particular policy target fits the pro- applying policy tools to specific target groups, file of a racial-group’s reputation. policy actors must rely on salient social classi- The RCM offers a clear basis for expectations fications and group reputations; without such regarding race and sanction decisions. We con- classifications, actors could not bring coher- ceptualize sanctions as tools for motivating ence to a complex social world or determine clients, stimulating work efforts, and enforcing appropriate action. responsible behavior. Accordingly, case man- Policy approaches that work for one group agers should be more likely to apply sanctions might fail for another. Thus, attempts to choose when clients are perceived as less motivated and effective policy actions inevitably depend on responsible—that is, when clients are perceived beliefs about “what kind” of group or individ- as needing a stronger push to follow rules and to ual one is seeking to influence. The first prem- achieve welfare-to-work goals. Client race should ise of the RCM holds that policy actors rely on thus affect sanction decisions to the extent that social categories to make complex target groups group reputations suggest differences in moti- more interpretable and, when making policy vation, work effort, and responsibility. Race- choices, draw on social-group reputations as based sanction disparities should increase when proxies for more detailed information about the contrast between reputations is larger; dis- these targets. parities should be weaker and less consistent 2. When racial minorities are salient in a pol- when the contrast is smaller. Finally, race-based icy context, race will be more likely to provide disparities should be more likely to emerge when a salient basis for social classification of targets proximate cues link clients to group reputations and, hence, to signify target differences per- in ways that highlight policy-relevant contrasts. ceived as relevant to the accomplishment of Like other students of implicit racism, we policy goals. are particularly interested in the potential for dis- The salience of race varies across policy crediting markers to cue group reputations in domains, time periods, and political jurisdic- ways that produce racial disparities (Quillian tions. All else being equal, we expect race to 2008). Two individuals seen as belonging to a 402 AMERICAN SOCIOLOGICAL REVIEW single racial group may nonetheless be associated SURVEY EXPERIMENTAL DESIGN with quite different group images. Research on intersectionality, for example, emphasizes that the Our experimental data come from a Web-based meaning of any category of social identity will survey of Florida Welfare Transition (WT) case likely change when combined with another (e.g., managers with sanctioning authority. The Florida when woman is combined with black as opposed WT program relies extensively on sanctions to to white [see Crenshaw 1991; Hancock 2007]). enforce client compliance, producing one of the Likewise, social cognition research shows that highest sanction rates for any state in the nation perceivers tend to distinguish “subtypes” of racial (Botsko, Synder, and Leos-Urbel 2001). Indeed, groups based on additional characteristics (e.g., a recent analysis of sanctioning in Florida finds “ghetto blacks” versus “black businessmen”) that the sanction rate for an entering cohort aver- and to attribute global group traits to these sub- ages nearly 50 percent after 18 months (Fording, types to very different degrees (Devine and Baker Soss, and Schram 2007). Florida is also one of 1991; Richards and Hewstone 2001). As a result, the most racially diverse states in the country, with race-of-target effects are often contingent on sizeable black and Hispanic populations, and the additional characteristics that strengthen or weak- state’s TANF population displays even more en an individual’s connection to a racial group’s diversity. Between January 2000 and March prevailing reputation. 2004, 36.2 percent of TANF adults were black, Eberhardt and colleagues (2006), for exam- 33.7 percent were white (non-Hispanic), and ple, find that black defendants convicted of 28.5 percent were Hispanic. Given its reliance on killing white victims are more likely to receive sanctions and its diverse client population, Florida the death penalty if they are perceived as hav- is an ideal state for examining the role of race in 1 ing a “stereotypically black appearance.” the sanctioning process. Likewise, Pager (2007) finds that black job To ensure anonymity for respondents, the applicants are significantly more disadvantaged Agency for Workforce Innovation (AWI) dis- than their white counterparts by having a felony tributed the survey link through e-mail to its 24 conviction on their records. Conversely, in their Regional Workforce Boards (RWBs) for sub- 2 study of black political candidates, Valentino, sequent distribution to caseworkers via e-mail. Hutchings, and White (2002:86) find that “when Case managers completed the surveys during a 3 the black racial cues are stereotype-inconsistent, two-week period at the end of 2006. Prioritizing the relationship between racial attitudes and the vote disappears.|.|.|. The effect emerges only when the pairing of the visuals with the narra- 1 All states must establish formal procedures to tive subtly reinforces negative stereotypes in ensure fairness when sanctioning clients, and the mind of the viewer.” Florida’s process is quite similar to that of other The RCM predicts that when minority clients states. For a detailed description of Florida’s sanction policy, see Appendix; for background on Florida’s possess discrediting traits consistent with minor- sanctioning practices, see Fording and colleagues ity stereotypes, they will be sanctioned signifi- (2007). cantly more often than (1) clients from all racial 2 AWI requested that the survey be distributed via groups who lack the discrediting trait and (2) an official electronic memorandum to RWBs. To white clients who share the discrediting trait. meet this request, we used a Web-based survey rather The RCM also predicts that this effect will be than a mailed, paper survey. RWBs were asked to for- greater for blacks than for Latinos because group ward the survey request to their contract agencies so stereotypes regarding work effort and personal they could encourage case managers to participate. responsibility are more negative in the case of This procedure was designed to motivate participa- blacks (Fox 2004; Gilens 1999). To test these tion without case managers feeling any direct pres- hypotheses, we make use of two client traits: one sure from the state government. The authors provided no additional incentives to promote participation in evokes the image of single mothers having chil- the survey. dren while engaging in long-term welfare 3 All caseworkers shared a survey password to dependency (see Hancock 2004); the other ensure anonymous completion and return. They were evokes images of a preference for living off wel- sent two follow-up requests for their participation, fare rather than pursuing the hard work of paid which extended the time period. We garnered no employment (see Gilens 1999). additional responses after the last follow-up request. DECIDING TO DISCIPLINE 403

Table 1. Distribution of Responses by Region

Region Name (Number) Frequency Percent Workforce Central Florida (12) 27 26.0 First Coast Workforce Development, Inc. (8) 22 21.2 Hillsborough County Workforce Board (15) 16 15.4 Pinellas Workforce Development Board (14) 11 10.6 Pasco-Hernando Jobs & Education Partnership Regional Board, Inc. (16) 6 5.8 Brevard Workforce Development Board, Inc. (13) 5 4.8 Southwest Florida Workforce Development Board (24) 5 4.8 Big Bend Jobs and Education Council, Inc. (5) 3 2.9 Citrus Levy Marion Workforce Development Board (10) 2 1.9 Polk County Workforce Development Board, Inc. (17) 2 1.9 North Florida Workforce Development Board (6) 2 1.9 Gulf Coast Workforce Development Board (4) 1 .96 Workforce Development Board of the Treasure Coast (20) 1 .96 Palm Beach County Workforce Development Board (21) 1 .96 Total 104 100.0 respondent anonymity limits our ability to deter- a sample well suited to the needs of our exper- mine whether all WT program officers passed imental design.5 Moreover, by triangulating our along the survey as requested, as well as the experimental results with an analysis of extent to which various regions participated in statewide administrative data, we offer an inde- the survey. We do, however, have regional pendent safeguard against any biases arising response data for the subset of case managers from nonparticipation in the survey. who elected to identify themselves by region, as Our analysis is based on two 2 2 experi- presented in Table 1. This table suggests that ments embedded in the survey, each of which responses were spread across a number of presented case managers with a vignette and regions but clustered in several of the state’s larg- asked them to decide whether to impose a sanc- er regions. tion. The 2 2 design includes variation on race We received survey responses from 144 and a discrediting social marker. Each vignette TANF case managers, the vast majority of portrays a hypothetical TANF participant who whom responded to our vignettes.4 Although has arguably fallen out of compliance with pro- Florida officials were unable to provide a pre- gram requirements. (For a description of rele- cise number for the overall population of WT vant rules and procedures in the WT program, case managers, estimates ran from 200 to 250. see Appendix.) For the racial dimension of the This suggests a rough estimate of between 58 2 2 design, each vignette makes use of a pro- and 72 percent for our response rate. Certain cedure developed by Bertrand and Mullainathan items in the questionnaire, however, including (2003), who show that by randomly assigning race (N = 98), political party affiliation (N = “black-sounding” names and “white-sounding” 103), and recent sanctioning behavior (N = names to a set of identical resumes, they can sig- 108), yielded a larger number of nonresponses nificantly influence the rate at which employ- than most. Table 2 provides a demographic pro- ers contact a fictitious job-seeker. Adapting this file of our survey respondents. Although procedure, we randomly assigned the client response patterns suggest that some caution should be used in generalizing from this sam- ple, the respondents are quite diverse and offer 5 Tests on the full sample, including respondents who did not indicate their race and other personal characteristics, provided baseline responses for our 4 A total of 137 caseworkers responded to Vignette experimental results that are consistent with our mul- 1 and 131 to Vignette 2. tivariate models. 404 AMERICAN SOCIOLOGICAL REVIEW

Table 2. Respondent Characteristics

Respondent Characteristics Percent of Sample Sex (N = 114) —Male 21.1 —Female 78.9 Race/Ethnicity (N = 98) —African American / Black 34.7 —European American / White 44.9 —Hispanic American / Latino 18.4 —Other 2.0 Educational Level (N = 115) —High school diploma 7.8 —Some college or trade school 33.0 —4-year college degree 37.4 —Some graduate school 13.0 —Graduate degree 8.7 Marital Status (N = 115) —Married 57.4 —Divorced/separated/widowed 20.9 —Single, never married 15.7 —Unmarried couple living together 6.1 Political Party Affiliation (N = 103) —Democrat or Independent Democrat 60.2 —Independent 11.7 —Republican or Independent Republican 24.3 —Other 3.9 Religious Attendance (N = 113) —Weekly 33.6 —At least once a month 23.9 —A few times a year 31.0 —Never 11.5 Mean years of welfare services experience (N = 143) 7.0 Note: N = number of caseworkers who responded to survey item.

described in Vignette 1 either a Hispanic-sound- We based our selection of these client traits on ing name or a white-sounding name. Similarly, our field interviews, which revealed substan- we assigned the client described in Vignette 2 tial caseworker attention to these two client either a black-sounding name or a white-sound- “types”: the young mother of multiple children ing name.6 and the repeat recipient who has been sanc- The second dimension of the 2 2 experi- tioned in the past. We present the vignettes ments manipulates client markers that are com- below with our experimental manipulations monly associated with (1) images of bracketed.7 undeserving welfare clients and (2) negative stereotypes of minority racial/ethnic groups. sounding names include Sonya Perez, Maria Rodriguez, and Luisa Alvarez. To test for name-spe- cific effects, we analyzed responses within each “race condition” to search for significant differences 6 To guard against confounding effects that might associated with each name. We found no significant arise from the use of a specific name, we randomly differences and thus treat all racial name cues as assigned one of three names for each group in each equivalent. vignette. White-sounding names include Sarah Walsh, 7 To minimize the amount of bias potentially intro- Emily O’Brien, and Meredith McCarthy; African duced by subsequent questions on sanctioning prac- American-sounding names include Lakisha Williams, tices, the two vignettes were presented near the Aisha Jackson, and Tamika Jones; and Hispanic- beginning of the survey. We tested for order effects, DECIDING TO DISCIPLINE 405

VIGNETTE 1 tifies the client as either white or Hispanic, a contrast associated with smaller stereotype [Emily O’Brien/Sonya Perez] is a 28-year-old differences than the black–white client con- single mother with [one child age 7 / four chil- trast used in Vignette 2. The client traits used dren ages 7 to 11, and who is currently in her in the two vignettes reinforce this difference. fourth month of pregnancy]. She entered the Vignette 1 focuses on childcare instability, Welfare Transition program six months ago, which is a “normal” rather than “deviant” after leaving her job as a cashier at a neigh- problem for women moving into employment borhood grocery store where she had worked (Loprest 2002). We expect the attribution of for nine months. Emily was recently reported numerous children here to cue negative for being absent for a week from her assign- stereotypes related to sexuality and repro- ment for community service work experience. duction, but this feature of the vignette also Immediately after hearing that Emily had not indicates that a comparatively sympathetic shown up for a week of work, Emily’s case- group (i.e., children) may suffer hardship as worker mailed a Notice of Failure to Participate a result of the sanction. As a result, one might (Form 2290) and phoned her to ask why she expect this cue to produce ambivalence about had missed her assignment. Emily was not sanctioning the family. The marker in Vignette home when the caseworker called. However, 2 is far less equivocal. By stating that the when she responded to the 2290 three days client has previously drawn a sanction, we later, she said she no longer trusted the person simultaneously provide case managers with who was looking after her [child/children], and two pieces of information that might cue per- she did not want to go back to work until she ceptions of welfare dependency and resistance found a new childcare provider. Emily returned to work the next day. to achieving self-sufficiency: the client is a repeat recipient who has returned to the pro- gram and has a record of at least one previ- VIGNETTE 2 ous failure to comply with welfare-to-work 9 [Emily O’Brien/Lakisha Williams] is a 26- rules. year-old single mother with two children. She Finally, the two narratives differ in relation to has been in the Welfare Transition program for sanction procedures. In Vignette 1, the client was five months. Lakisha was recently reported reported for a week’s absence from her work for failing to show up for a job interview that experience assignment. Although she was not had been scheduled for her with a local house- home when the caseworker called, she respond- cleaning service. Immediately after hearing ed to the mailed 2290 form well within the 10- about the missed interview, Lakisha’s case- day period allotted and reestablished compliance worker mailed a Notice of Failure to by returning to work the next day. According to Participate (2290) and phoned her to ask why both the sanctioning rules and the WT Sanction she had not shown up. Lakisha said she had skipped the interview because she had heard ple occasions in sanctions training for case man- that a better position might open up next agers. We developed the vignettes in consultation month with a home health agency. [She had with AWI staff responsible for this training. The sur- been sanctioned two months earlier for fail- vey/experiment was pilot tested with other AWI staff ure to complete her hours for digital divide.] who had prior experience as frontline case managers. 9 The two experiments offer a more likely Under Florida’s WT program, a record of a sanc- tion two months prior to the current failure to com- and a less likely case for finding effects on 8 ply should not itself increase a client’s jeopardy of sanction decision-making. Vignette 1 iden- receiving a sanction for the present offense (see Appendix). The client in the second vignette, how- ever, has a prior sanction and therefore faces a length- specifically, whether sanctioning decisions in Vignette ier penalty (one month) for noncompliance than 1 predict sanctioning decisions in Vignette 2, and find would a client without a prior sanction (10 days). It no evidence of such effects. is possible that a case manager might be less willing 8 We developed the questionnaire, including the to impose a sanction knowing that a client faces a vignettes, after extensive study of the sanctions longer penalty and therefore a greater amount of process in Florida, including participation on multi- time without cash assistance. 406 AMERICAN SOCIOLOGICAL REVIEW

Flow Chart (see Appendix), the pre-penalty black-named clients were slightly more like- phase for this client should end with compli- ly than the white-named clients to be recom- ance following her return to work. In other mended for sanction (82.3 versus 76.8 words, based solely on the vignette, one can- percent, p = .22). Consistent with our hypoth- not say that a sanction is the appropriate esis, the racial disparities widen with the pres- response. By contrast, although the client in ence of a discrediting marker in each Vignette 2 responded to the case manager’s experiment. For Vignette 1, the Hispanic- phone call, her reason for not complying with named clients who are pregnant and have four her welfare contract clearly fails to meet children were likely to be sanctioned 40 per- requirements for a “good cause” exemption. 10 cent of the time, while the non-Hispanic The two vignettes present us with an oppor- white-named clients were slightly less likely tunity to test the RCM in strikingly different to be sanctioned at 27.3 percent of the time cases. Each vignette was followed by a ques- (p = .11). For Vignette 2, the black-named tion asking respondents to indicate on a four- clients with a prior sanction were likely to be point scale whether they strongly favor, sanctioned 93.9 percent of the time, com- somewhat favor, somewhat oppose, or strong- pared with 77.4 percent of the time for the ly oppose requesting a sanction for the client previously sanctioned white-named clients and situation described.11 (p < .05). These simple descriptive results, however, fail to take account of possible dif- SURVEY EXPERIMENTAL ANALYSIS: ferences between managers assigned to each RESULTS condition of the experiment. To do so, we include relevant covariates in a larger multi- As anticipated, fewer caseworkers recom- variate analysis. mended sanctioning the client in Vignette 1 Our multivariate models include dichoto- (34 percent) than in Vignette 2 (79 percent). mous variables representing three of the four At the same time, these overall results strong- experimental conditions (the white client with ly confirm our expectation of variation in a “more deserving” trait serves as the base- caseworker judgment: roughly a third of case- line). The models also include measures of workers decided that a sanction was war- selected case manager characteristics that, ranted in Vignette 1, despite the fact that WT based on literature on welfare casework, we program rules seem to suggest otherwise, and have reason to believe affect a willingness to 21 percent opted not to sanction the client in impose sanctions: work experience, religios- Vignette 2, despite a clear violation of the ity, education level, partisan identification, rules. marital status, and racial identity (Dias and Our baseline results indicate differences Maynard-Moody 2007; Gooden 2003; in how different clients are treated. The Watkins-Hayes 2009). We hypothesize that Hispanic-named clients were more likely to case managers with greater experience in the be recommended for sanctioning than were social welfare field are likely to have either the white-named clients in Vignette 1 (40 ver- worked under the earlier, more permissive sus 28.6 percent, p < .10); in Vignette 2, the system of welfare or to have witnessed the negative effects of sanctioning; they are there- fore less likely to impose sanctions. Case 10 Although failure to provide evidence of the managers with more experience may also be childcare problem, or the children being too old more committed to supporting clients and under Florida’s childcare exemption policy, may pre- thus less likely to support sanctioning. Highly clude a “good cause” exemption, Vignette 1 intro- religious case managers are likely to have duces the idea that “good cause” might apply here greater commitment to enforcing basic work because the mother provided some evidence of her and family values and thus are possibly more childcare problem and came back into compliance within the time allowed. likely to impose sanctions. More educated 11 In Florida, sanctions are all of one type—finan- case managers might be less likely to impose cial penalties for failure to comply with program sanctions because they are apt to be more rules. Penalties increase with each infraction (see informed about their effects. Democrats are Appendix). probably less likely to adhere to the new wel- DECIDING TO DISCIPLINE 407

Table 3. Analysis of Vignette Experiments with Caseworker Characteristics as Covariates

Independent Variables Vignette 1 Vignette 2 Vignette Condition —White client, marked .501 .367 (.60) (.51) —Minority client, unmarked 1.383†–.075 (1.50) (–.11) —Minority client, marked 1.693* 2.599** (1.96) (2.28) Caseworker Characteristics —Experience –1.573** –1.632* (–2.73) (–1.98) —Religiosity –.314 .018 (–.56) (.03) —College education .134 .424 (.24) (.73) —Democrat .567 –.127 (.99) (–.22) —Married .903 –.152 (1.48) (–.25) —Black or Hispanic .301 .405 (.48) (.63) LR ␹2 16.90* 16.90* Log likelihood –46.304 –41.482 Pseudo R2 .154 .169 N9594 Note: Entries are coefficients followed by z-scores in parentheses. For Vignette 1, the racial minority is Hispanic and the marked condition is “four children and pregnant,” as opposed to one child. For Vignette 2, the racial minority is black and the marked condition is possession of a prior WT sanction, as opposed to no mention of a prior participation spell. The number of cases for each model is lower than the overall sample due to missing data for selected covariates. † p ≤ .10; * p ≤ .05; ** p ≤ .01 (one-tailed tests). fare regime’s use of sanctions to enforce client sanctions. No other covariate achieves statisti- compliance. Married case managers might be cal significance, and we obtained no significant more likely to support sanctions, perhaps as an results in further specification tests of case- expression of a commitment to upholding tra- worker characteristics not included here. ditional values. We include racial identity based Caseworkers with more than two years expe- on the hypothesis that white case managers rience are significantly less likely than their might be more willing to sanction nonwhites. less experienced counterparts to choose a sanc- Furthermore, nonwhite case managers might tion in both Vignette 1 (p < .01) and Vignette 2 feel they lack the “race privilege” to sanction (p < .05). Several factors may explain why more whites, while also perceiving special obliga- experienced case managers are less likely to tions to sanction black clients out of grounds of impose a sanction. For example, case managers loyalty to the race or a desire to teach clients of with more experience are more likely to have the same race the need to be compliant with been trained in an earlier era of welfare provi- established norms. sion that placed less emphasis on sanctioning Table 3 presents the results of logit models for (Aid to Families with Dependent Children prior both vignettes. Only one of the covariates to 1996, or Florida’s initial reform program emerges as statistically significant, and it does Work and Gain Economic Self-Sufficiency so consistently across experiments. Case man- [WAGES] 1996 to 2000). Alternatively, if case agers with more experience (defined as those managers learn over time that sanctions have with more than two years of welfare services negative consequences for clients or for per- experience) emerge here as less likely to impose formance ratings, then more experienced case 408 AMERICAN SOCIOLOGICAL REVIEW managers might become more reluctant to applied is significantly lower than what we impose sanctions. Another possibility is that observe when the client is identified as a black experienced case managers develop greater woman with a prior sanction on her record. knowledge of how to work with clients and, To put these results into perspective, we can hence, resort to sanctions less quickly than do calculate the predicted probability of being sanc- novice case managers. Our data do not allow us tioned for each of the four client conditions we to arbitrate among these alternatives.12 examine in each vignette by holding all the other The key findings in Table 3 are indicated by variables in the model at their median and allow- the coefficients for our experimental conditions. ing the variable in question to vary. Figure 1 The results are striking for their consistency, provides the predicted probabilities for each of especially given the considerable differences the four types of clients for both vignettes. Based between our two vignettes. In the first, we find on our model for the first vignette, we find that that case managers are no more likely to sanc- a Latina mother who is pregnant and already has tion the white client with multiple children than four children has a .47 predicted probability of to sanction the white client with one child. We being sanctioned, while a white mother in the find borderline results suggesting that case man- same condition has only a .21 predicted proba- agers may be more likely to sanction the bility of being sanctioned, and a white mother Hispanic client with one child than to sanction with only one child has only a .14 probability of the white client with one child (one-tailed test, being sanctioned. For Vignette 2, a black client p = .067). By contrast, we find that a Hispanic with a prior sanction has a .97 predicted proba- woman with multiple children is significantly bility of being sanctioned, compared with .75 for more likely to be sanctioned compared with a a white client with a prior sanction and .67 for white woman with one child. a white client without a prior sanction. The results for Vignette 2 tell a similar story, The consistency of findings across two ran- only in stronger form. Here, we find no dis- dom-assignment experiments, each with very cernible differences in the odds that a case man- different conditions, is noteworthy. The key lim- ager will sanction a white woman with no prior itation of the evidence, however, is that these sanction, a black woman with no prior sanction, findings, like most studies of implicit racism, are or a white woman with a prior sanction. Sanction based on hypothetical scenarios (Duster 2008). decisions are, from a statistical perspective, When case managers responded to these invariant across these conditions. When a prior vignettes, they were not confronted with a real sanction is attributed to a black woman, howev- person: they did not have a detailed case file in er, we find a large and statistically significant hand, they did not have to worry about effects on increase in the possibility of being sanctioned, their performance numbers, and they did not compared with a white women without a prior have to contemplate real material hardships that sanction.13 Indeed, under every other combina- might result. To bring our empirical analysis tion of race and sanction history in our experi- into line with real-world conditions, we must turn ment, the probability of a new sanction being to administrative data generated by the Florida WT program itself. In doing so, we lose certainty about whether clients with different character- 12 In alternate specifications of our models, we find istics have equivalent cases and must rely on an no significant interactions between case manager imperfect process of specifying control vari- experience and client race. The effects of case man- ables. In return, however, we gain the ability to ager experience should thus be seen as independent triangulate our survey-experimental findings of the racial patterns presented below. 13 with data that bear a closer relationship to the real We tested our models including interactions world of administrative practice. that combine the race of the caseworker with the race of the client to see if the results could be better explained by intergroup hostility between case- ADMINISTRATIVE DATA: EVENT workers and clients, rather than the logic of the RCM. HISTORY ANALYSIS The interaction terms proved insignificant, suggest- ing the results are not reflective of intergroup dif- The Florida Department of Children and Families ferences between the caseworkers and the clients (DCF) provided the data for these analyses, which and that the RCM offers a better explanation. consist of individual-level records for all adults DECIDING TO DISCIPLINE 409

Vignette 1 Vignette 2 Figure 1. Predicted Probabilities by Race and Condition who received TANF in Florida between January 1 as closely as possible, we then restrict our 2000 and April 2004. Our data set consists of sample to unmarried, female clients who are monthly records for reporting case outcomes either white (non-Hispanic) or Hispanic and and the characteristics of TANF clients and their who have either one or four children. The families. Our analytic strategy is to replicate the dependent variable for our analysis is a dichoto- experimental vignettes as closely as possible mous variable that takes on a value of 1 in the by estimating an event history model of sanc- month that a client is sanctioned. We follow tion initiation, focusing on the effect of a client’s each of the 24 cohorts in our sample for up to race or ethnicity and its interaction with the a maximum of 12 consecutive months, ending stigmatizing marker used in each experimental our observation at the spell’s termination or the vignette. We begin by examining the joint effects 12-month mark, whichever comes first. Clients of ethnic identity and family size tested in who exit for reasons other than a sanction, or Vignette 1. We then turn to the joint effects of who are not sanctioned by the 12th month of the racial status and sanction history tested in spell, are treated as right-censored. For our first Vignette 2. analysis, we restrict our attention to the first TANF spell for each individual during this peri- TRIANGULATING VIGNETTE 1: od; we define a spell as continuous months of ETHNICITY AND FAMILY SIZE For our first analysis, we construct a sample of all new adult clients who entered TANF during who have spent at least 12 continuous months with- the 24-month period from January 2001 through out TANF benefits. This precludes the inclusion of December 2002.14 In an effort to match Vignette clients entering TANF during 2000. In addition, we wish to observe each client cohort for up to 12 months after entering TANF. This forces us to exclude clients 14 Our selection of cohorts is limited by two fac- who entered TANF during the last year of our obser- tors. First, we define “new” TANF clients as those vation period. 410 AMERICAN SOCIOLOGICAL REVIEW

TANF receipt.15 As defined, and accounting for percent Hispanic,19 and several measures of a small percentage of cases for which values of employment conditions: the county unemploy- some variables are missing, our total sample ment rate, the county poverty rate, the level of size exceeds 6,000 women and nearly 20,000 urbanization as measured by county popula- person-month observations. tion, and the annual local wage in food ser- We estimate our model using a Cox propor- vice/drinking establishments.20 Finally, we tional hazards approach. The advantage of the include a measure of the county TANF caseload, Cox model is that it allows for flexible, non- expressed as a proportion of the county popu- parametric estimation of the baseline hazard, or lation.21 what we might think of as the effect of spell Table 4 shows the results for our event his- duration on the probability of sanction (Box- tory model. Column I presents the results for 16 Steffensmeier and Jones 2004). Our primary a model that estimates the additive effect of interest lies in the effect of a client’s ethnicity ethnicity through the inclusion of a single and changes in this effect as we move from dichotomous variable (Hispanic) measuring clients with only one child to clients with four whether a client is white or Hispanic. Column children. To test this interactive hypothesis, we II presents results for a model that tests for an divide our sample into four groups that paral- interactive effect between ethnicity and the lel the experimental groups in Vignette 1. number of children using the indicator vari- Based on past research on sanctions and wel- ables described above. For each variable in fare implementation (Hasenfeld et al. 2004; Kalil et al. 2002; Wu et al. 2006), we include a number of independent variables to control for variation in clients’ individual and community 19 Previous studies often find that racial context has characteristics.17 At the individual level, we a significant effect on racially relevant policy out- include variables measuring a client’s citizenship comes. The “threat” hypothesis (Blalock 1967; Key status, her age, the age of the youngest child in 1949) suggests that the higher the percentage of non- the TANF family, and two indicators of human whites, the greater the support for more punitive capital (income and education). We also control policies. The “contact” hypothesis (Alport 1954; Fox 2004) indicates that the higher the percentage of for a variety of community conditions, includ- 18 nonwhites, the lower the support for punitive policies. ing local conservatism, percent black, and There are also possible effects from an increased proportion of the population nonwhite enhancing minority political power (Keech [1968] 1981). 15 Based on findings from our field research, we Because there is reason to suspect that any of these do not include the first two months of the first TANF effects might exist (see Keiser, Mueser, and Choi spell in our analysis. In interviews at all levels of the 2004), we test for effects of community racial com- WT program, officials report that sanctions record- position by including the percentage of the county ed in the first months of a spell often represent a form population that is Hispanic and black, respectively. of “self-sanctioning” that is distinguishable from 20 Where employment opportunities are relative- “true sanctioning” decisions made by case managers. ly numerous and attractive, TANF clients may be In this scenario, an applicant with some alternative more likely to work enough hours to avoid falling out income options enters the official rolls, begins to of compliance with TANF rules. Alternatively, local receive assistance, but then does not return to the local labor market conditions may influence the sanction provider after learning what will be required of her decisions of case managers, who may be less inclined and how much cash aid she will receive in return. to sanction clients when job opportunities are less 16 We have replicated our findings using other numerous or less attractive. estimation methods as well, including parametric 21 As the caseload size increases, administrative methods (Weibull) and a discrete-time (logit) model. pressures to reduce caseloads should result in an 17 See Table A1 in the appendix for variable defi- increase in sanctioning. Additionally, the heightened nitions and sources. burden of more cases may increase the chances that 18 Several studies find that local policy imple- caseworkers will rely on race-relevant heuristics. mentation is influenced by the local political envi- Alternatively, as caseload size increases, if the num- ronment (Cho et al. 2005; Weissert 2000). Based on ber of case managers remains fixed, individual case these studies, we include a measure of local politi- managers may have less time to closely monitor cal ideology, which we expect will be positively relat- TANF clients for violations of rules, thus resulting ed to sanction initiation. in a lower rate of sanctioning. DECIDING TO DISCIPLINE 411

Table 4. Cox Proportional Hazard Models of Effects of Minority Status and Number of Children on Sanction Initiation

III Independent Variables Additive Model Interactive Model Individual Characteristics —Hispanic client .960 —White client and one child [Reference] —White client and four children .981 —Hispanic client and one child .955 —Hispanic client and four children .982 —Age of client .980** .980** —Age of youngest child 1.013 1.013 —Citizenship status (1 = citizen) 1.276 1.274 —Education (reference = > high school) ——Less than high school education 1.371** 1.370** ——High school education 1.068 1.068 —Income (in 1000s) 1.067** 1.067**

Community Characteristics —Local conservatism 1.085 1.085 —Percent black .997 .997 —Percent Hispanic .987** .987** —Annual wage – food service/drinking places .976 .976 —Unemployment rate t – 1 1.029 1.029 —Poverty rate 1.017 1.017 —Population (in millions) 1.281** 1.281** —TANF caseload t – 1 .911 .911

Number of subjects 6,214 6,214 Number of failures 1,792 1,792 Time at risk (person–months) 19,798 19,798 Note: The sample for this analysis consists of all new TANF clients (single-parent, female, Hispanic or white) who entered TANF from January 2001 through December 2002. All clients are observed for a maximum of 12 months (clients who exit without being sanctioned, or who were sanctioned after 12 months, are treated as cen- sored). Cell entries are hazard ratios, with p-values based on robust standard errors (adjusted for error clustering at the county level). * p < .05; ** p < .01 (two-tailed tests). our models, we report the estimated hazard Koraleck 2000; Mancuso and Lindler 2001; ratio, which reflects the proportional change Wu et al. 2006).22 in the risk of sanction given a one-unit Moving to our main hypotheses of interest, increase in the independent variable of inter- we find no evidence in the administrative data est. Most of the control variables in the model to suggest that Hispanic clients are sanctioned perform as expected. The risk of being sanc- at a greater rate than white clients. This is true tioned is lower among older and more edu- regardless of whether we examine the additive cated clients and among those who reside in effect of ethnicity displayed in column I or its a community with a large Hispanic popula- conditional effect in column II. Our first analy- tion. Clients are more likely to be sanctioned sis of administrative data thus proves to be if they have older children or reside in heav- ily populated counties. These results are large- ly consistent with the findings of past studies 22 We also tested for whether black clients with four of sanctioning (Born, Caudill, and Cordero children were more likely to be sanctioned than 1999; Fording et al. 2007; Hasenfeld et al. whites with one child and the results are not signif- 2004; Kalil et al. 2002; Keiser et al. 2004; icant. 412 AMERICAN SOCIOLOGICAL REVIEW inconsistent with the findings for Vignette 1 we are trying to explain (sanctioning). Because from our survey of case managers. this nonrandom selection may introduce statis- tical bias, we must control for the selection TRIANGULATING VIGNETTE 2: RACE AND process that brings some (but not all) first-spell clients into a second spell. To do so, we rely on PRIOR SANCTIONS an estimator introduced by Boehmke, Morey, We now move to an analysis of administrative and Shannon (2006) for continuous-time event- data based on Vignette 2, which examined the history models with sample selection. The pro- joint effects of client race (black/white) and cedure relies on full information maximum sanction history. Once again, we limit our sam- likelihood to simultaneously model the selection ple to clients with the characteristics described and event history processes. Like the Heckman in the vignette—unmarried female clients who model for continuous outcomes, we first esti- are either black or white (non-Hispanic). We also mate a binary model of the selection process— rely on the same general research design by a client’s return to TANF after having exited a estimating an event history model of sanction first spell.24 We then use the information gar- initiation. The narrative in Vignette 2, howev- nered from the selection model to correct for er, presents us with two additional complications selection bias in the event history model. for an analysis of administrative data. First, as Because the estimation procedure is limited to the stigmatizing condition in Vignette 2 is a parametric event history models, we use the prior sanction, we must go beyond the first spell Weibull distribution to model duration depen- to examine sanctioning outcomes for clients dence (i.e., the effect of time on sanction initi- with a prior history of TANF participation. We ation).25 therefore rely on a sample consisting solely of Once again, we divide our sample into groups TANF clients participating in their second analogous to the four experimental groups fea- spell.23 Specifically, the sample consists of all tured in Vignette 2. We include indicators for TANF clients (unmarried, female, black or three of the groups in the model, excluding white) who entered TANF for the first time white and no prior sanction as the reference between January 2001 and December 2002, group. Because our theory suggests that whites and who returned to TANF for a second spell with no prior sanction should be subject to the during this same period. As defined, and lowest rate of sanction, we expect that the coef- accounting for some missing data, our sample ficients representing each of the three indicator consists of approximately 19,000 second spell variables included in the model will be positive clients, approximately 37 percent of whom (and the associated hazard ratios > 1.0). In addi- experienced a sanction during their second spell. tion, we expect that the variable representing A second complicating factor arises due to the blacks with a prior sanction (black and prior possibility of sample selection bias. That is, the sanction) will display the largest coefficient. factors that cause TANF recipients to return to Finally, we also include several individual and the program for a second spell (and thus enter community-level controls: age of client, age of our sample) may also be related to the outcome youngest child, education, income, local con- servatism, percent black, annual wage, poverty rate, and TANF caseload.26 23 While our data allow us to examine sanctioning outcomes beyond the second spell, we limit our analysis to the second spell for at least two reasons. 24 We do not present this model in the text, but see First, because our observation period is fixed, each Table A2 in the Appendix. successive spell necessarily increases the number of 25 The Weibull distribution allows for a flexible right-censored observations. Second, the receipt of a specification of the baseline hazard rate and is appro- third sanction can result in a TANF client being priate when the baseline hazard rate is monotonical- barred from TANF participation for three months, ly increasing, monotonically decreasing, or flat over even if the client comes into compliance with TANF time. Based on various diagnostics, we are satisfied rules. We suspect that this may alter case managers’ that the Weibull model is appropriate for our data. decision-making processes in ways that render the 26 We do not include citizenship status, number of sanction decision less comparable to the decision children, percent Hispanic, or population, because prompted by Vignette 2. these variables proved to be highly insignificant in DECIDING TO DISCIPLINE 413

Table 5. Weibull Selection Model of Effects of Minority Status and Sanction History on Sanction Initiation during Second TANF Spell

III Independent Variables Additive Model Interactive Model Individual Characteristics —Black client 1.190** —Prior sanction 1.091* —White client and no prior sanction [Reference] —White client and prior sanction 1.032 —Black client and no prior sanction 1.145* —Black client and prior sanction 1.288** —Age of client .968** .968** —Age of youngest child 1.012 1.012 —Education (in years) .979** .979** —Income (in 1000s) .975 .976

Community Characteristics —Local conservatism 1.029 1.028 —Percent black .992* .992* —Annual wage – food service/drinking places .973 .973 —Poverty rate 1.002 1.002 —TANF caseload .887* .887*

Rho (error correlation) –.223** –.223** Total N 40,891 40,891 Uncensored N (returning for 2nd spell) 18,827 18,827 Note: The sample for this model consists of all TANF clients (single-parent, female, black or white) who entered TANF from January 2001 through December 2002 and returned for a second TANF spell during this same obser- vation period. All clients are observed during the 2nd spell until they are sanctioned or they exit TANF for other reasons. Clients who exited TANF without being sanctioned, or whose second spell continued beyond the close of our observation window (April 2004), were treated as right-censored observations in the Weibull model. The model was estimated in Stata 10.0 using the DURSEL procedure (Boehmke 2005). Cell entries are hazard ratios, with p-values based on robust standard errors (adjusted for error clustering at the county level). * p < .05; ** p < .01 (two-tailed tests).

Table 5 presents the results for our event- increased contact with the nonwhite popula- history model. The effects of the control vari- tion, it appears that a large minority presence ables largely conform to our expectations and within a community may offset any racial or eth- look similar to the first-spell results reported in nic biases in sanctioning that could occur with- Table 4. The risk of sanction is highest for in the implementation process. younger clients who are less educated, and the Moving to the results for our primary risk is lower among clients who live in counties hypotheses, we find strong corroboration of with relatively large black populations and high our experimental findings. The additive speci- TANF caseloads. The negative effect of per- fication in column I tests for the independent cent black is especially interesting given the effect of race. The hazard ratios reported here indicate that black clients are indeed signifi- fact that we found a similar effect for percent cantly more likely than white clients to be sanc- Hispanic for the sample that included Hispanic tioned, regardless of their sanction history. The clients. Whether through increased presence results in column II allow for a test of our inter- within the local welfare bureaucracy, outside active hypotheses and suggest that these two pressure on local welfare policymakers, or characteristics do interact as expected. Indeed, if we begin with the baseline risk of sanction for a white client who has no prior sanction, we find preliminary models. Their inclusion in the model that the positive effect of making this client does not affect the results reported in Table 5. black is significantly greater than the effect of 414 AMERICAN SOCIOLOGICAL REVIEW giving the white client a prior sanction. Further, managers with more than two years experience having a prior sanction does not seem to mat- were significantly less likely to impose sanctions ter much for white clients, but it matters a great in both experiments. Although we discussed deal for black clients. The hazard ratio repre- several possible explanations for this pattern, our senting the difference in risk of sanction between data do not allow us to distinguish among them, whites with and without a sanction is close to thus raising an important puzzle for future 1.0 and is far from being statistically significant research. (p = .63). By contrast, the risk of sanction for In both law and principle, welfare sanctions black clients with no prior sanction is approx- should be imposed as responses to client behav- imately 14 percent higher than that for white ior. In practice, however, we find that sanctions clients without a first-spell sanction (p < .05). are also used in response to client characteris- Among black clients with a prior sanction, this tics. Despite having identical case narratives, our risk is doubled; the risk of sanction for black first vignette finds that a pregnant Latina client clients with a prior sanction is 28 percent high- with four children is significantly more likely er during the second spell than the risk for white to be sanctioned than a white client with only clients with no prior sanction (p < .01). In sum- one child. In the second vignette, we find that mary, even though black clients with no prior a black client with a prior sanction is signifi- sanction are already at a higher risk than white cantly more likely to be sanctioned than a white clients of being sanctioned in the second spell, client with no prior sanction. The two vignettes the addition of a prior sanction increases black are quite different—in the racial/ethnic con- clients’risk of sanction to a significantly greater trast involved, the nature of the stereotypical degree than for white clients.27 trait, and their relation to program rules regard- ing sanctioning—but the results are largely the DISCUSSION AND CONCLUSIONS same. White clients in these experiments suffer no statistically discernible negative effects when This article confronts, more directly than past linked to characteristics that hold negative mean- research, the question of how and why race ings in the welfare-to-work context. As advo- influences sanctioning under welfare reform. To cates of administrative consistency might hope, do so, we apply a general model developed to case narratives elicit a stable pattern of respons- explain how racial classifications affect policy es from case managers, regardless of discredit- choices in diverse domains. The RCM provides ing attributes, when clients are white. Minority clear, testable predictions about where and when clients enjoy no such immunity: their odds of racial disparities are likely to emerge in the being sanctioned rise in the presence of dis- administration of welfare programs. Our exper- crediting markers, even when the details of their imental results support these predictions and case do not change a bit. are corroborated by administrative data indi- Our random-assignment vignettes offer a cating how real clients under welfare reform crisp test of causal effects, but this power is have actually been treated. The results converge purchased at some cost. In the survey setting, to provide striking evidence for both the utili- case managers are pulled out of their normal ty of the model and the enduring power of race organizational environment, client narratives in U.S. poverty governance. are reduced to a handful of details, and hypo- Our experiments randomly assigned case thetical sanction decisions involve no cost to managers to vignettes in which clients differed case manager time or recipient-family well in their race/ethnicity and in the possession of being. To close this gap, we triangulated our stereotype-consistent discrediting traits. Only experimental findings with an analysis of one caseworker characteristic emerged as a sig- administrative data, a source of evidence that nificant predictor of sanction decisions: case provides a weaker basis for causal inference but more faithfully reflects actual practices on the ground. In so doing, we test experimental 27 We tested in the administrative data whether evidence of implicit racial bias against evidence Hispanic clients with a prior sanction were more arising from the actual exercise of government likely than white clients to be sanctioned again and authority (Duster 2008). The results of this found this not to be the case. analysis do not support our experimental find- DECIDING TO DISCIPLINE 415 ings regarding the attribution of multiple chil- While Latinas might face other forms of dis- dren to a Hispanic woman. In the administrative crimination in the welfare system, including data, Hispanic clients are no more likely than language barriers, our evidence suggests that white clients to be sanctioned, and this null racial classification leads to greater disparities finding holds regardless of number of children. in sanctioning for blacks than it does for By contrast, we find strong support in the Hispanics. administrative data for our experimental results Our analysis does not directly test whether in Vignette 2. Among second-spell participants, sanctioning is influenced by racial animus, black clients with a prior sanction are more threat, or loyalty—as opposed to the more cog- likely than their white counterparts to be sanc- nitive dynamics emphasized by theories of tioned. White clients suffer no discernible implicit racism (Quillian 2008) and the RCM increase in their risk of sanction when they have (Soss et al. 2008). It is worth noting, however, a prior sanction, while black clients—who are that we do not find evidence in our experiments already more likely to be sanctioned than that white case managers differ from nonwhite whites—become significantly more likely to case managers in their sanction decisions. White be penalized when this discrediting marker case managers were no more likely to sanction appears on their record. clients overall and no more likely to be influ- Together, these findings offer powerful evi- enced by our experimental manipulations of dence that racial status and stereotype-consis- race/ethnicity and client traits. These findings tent traits interact to shape the allocation of are hard to square with an account that empha- punishment at the frontlines of welfare reform. sizes white ingroup loyalty or white animus The lone instance in which our findings do not toward blacks. They are far more consistent converge (the treatment of Hispanics in the with models emphasizing how racial classifi- experiment versus the administrative data) might cations operate in implicit ways—without con- suggest some inconsistency. Viewed in the con- scious racism—to generate racial disparities text of the RCM, however, it actually conforms (Quillian 2008). Race matters in more subtle to a predictable and repeated pattern. Prior ways than overt hostility or loyalty; race is built research indicates that the gap between Hispanic into the cognitive processes that provide the and white stereotypes is smaller than the gap foundation for decisions about how target between black and white stereotypes (Fox 2004; groups should be treated in welfare policy set- Gilens 1999). Accordingly, the RCM predicts tings (Schram 2005). that disparities will emerge more strongly and Sanctioning practices under welfare reform consistently in the latter case. are part of a larger turn toward disciplinary Indeed, we have found this precise pattern of poverty governance in the United States race-based disadvantage in studies of other (Wacquant 2001). In this context, it is impera- aspects of welfare reform (Fording et al. 2007; tive that social scientists begin to provide some Schram, Fording, and Soss 2008; Soss et al. insight into how disciplinary practices operate 2008). This article extends this body of evi- and how the state’s authority to punish may be dence by showing that African Americans are used in ways that deepen or ameliorate social distinctively vulnerable to the presence of dis- inequalities. TANF is ostensibly a race-neutral crediting social markers. In the present study, the public policy, but it is carried out today in ways stigma of deviant behavior attaches most strong- that allow preexisting racial stereotypes and ly to black clients, more weakly and less con- race-based disadvantages to produce large sistently to Hispanic clients, and not at all to cumulative disadvantages (Schram 2005, 2006). white clients. This pattern is further supported Our prior research finds that black TANF recip- by our analysis of administrative data showing ients, relative to their white counterparts, are that Hispanic clients—unlike black clients— more likely to participate in the toughest poli- are no more likely to be sanctioned when they cy regimes controlled at the most local levels possess a prior sanction on their record. This pat- (Soss et al. 2008). Within one such regime, tern closely resembles the results Pager (2007) Florida, we find that they are more likely to be reports in her landmark study of how race and sanctioned and their odds of being sanctioned markers of criminality interact to limit African are more likely to rise when they are associat- American men’s prospects in the labor market. ed with longer participation spells or participate 416 AMERICAN SOCIOLOGICAL REVIEW in conservative regions (Fording et al. 2007). Mawr College. Her doctoral dissertation examines While Florida may be distinctive in many ways, factors affecting childcare disruption and care-relat- studies comparing Florida with other states sug- ed interruptions in employment and, specifically, gest good reasons to think that the racial dynam- how such factors may be experienced and operate dif- ics we have uncovered are not unique to welfare ferentially by place. implementation in this locale (Shaw et al. 2006). The results presented in this article suggest that APPENDIX: SANCTIONING welfare sanctions, once imposed, become dis- PROCEDURES IN FLORIDA crediting markers that make black clients even Florida’s Welfare Transition program is admin- more vulnerable to a future sanction. Our find- istered by public, nonprofit, and for-profit ings reinforce the conclusion that policy choic- provider agencies under contract with the State’s es not only reflect but also create the elements 24 Regional Workforce Boards (RWBs). In a that underpin racial inequality in the U.S. wel- series of program guidelines issued in February fare system. Under cover of a policy that is of 2004, Florida’s Agency for Workforce officially race-neutral, welfare systems oper- Innovation (AWI) and the Department of ate in ways that reflect racial classifications, Children and Families (DCF) clarified the reproduce racial inequities, and call out for Florida statutes relating to work penalties and attention from both scholars and reformers. pre-penalty counseling, with one goal being to Sanford F. Schram teaches social theory and poli- “develop integrated and consistent procedures cy at the Graduate School of Social Work and Social to implement sanctions” (AWI FG 03-037, 1). Research, Bryn Mawr College. He has published According to these guidelines, a first occasion articles in the American Sociological Review, the of noncompliance with a work contract results American Political Science Review, the American in a full-family termination of temporary cash Journal of Political Science, and numerous other assistance for a minimum of 10 days or until the journals. His most recent book is Welfare Discipline: individual reestablishes compliance. Second Discourse, Governance, and Globalization (Temple and third instances of noncompliance are 2006). attached to periods of termination of one month Joe Soss is Cowles Professor for the Study of Public and three months, respectively. Sanctions that Service at the Hubert H. Humphrey Institute of Public remain in place for more than a 30-day period Affairs, University of Minnesota. His published arti- can be resolved only when the return to com- cles have appeared in American Political Science pliance has been documented and is accompa- Review, the American Journal of Political Science, the Journal of Politics, and other journals. He is the nied by a new Request for Assistance (RFA) and author of Unwanted Claims: The Politics of face-to-face interview (F.S. 445.024). Previous Participation in the U.S. Welfare System (Michigan sanctions may be “forgiven” for participants 2002) and coeditor of Race and the Politics of Welfare who are compliant with their welfare contracts Reform (Michigan 2003) and Remaking America: for a period of six months. Following a six- Democracy and Public Policy in an Age of Inequality month compliant period, any sanction levied (Sage 2007). against a participant is treated as a first occa- Richard C. Fording is Professor of Political Science sion (AWI FG 03-037; F.S. 414.065). at the University of Kentucky. He is also Associate The Florida Statutes allow for “good cause” Director of the University of Kentucky Center for exceptions to the sanctioning policies outlined Poverty Research. His published research has above. This may represent the point at which the appeared in the American Political Science Review, preferences and understandings of individual the American Journal of Political Science, the Journal local actors most clearly come into play (Pavetti of Politics, and other journals. He is coeditor of Race and the Politics of Welfare Reform (Michigan et al. 2003). Such exemptions may be granted 2003). The research analyzed in this article comes for instances of noncompliance related to child- from a larger research project: http://www.uky. care,28 the current or past effects of domestic edu/~rford/floridaproject.htm. The results from this project will be reported in a book tentatively titled “Disciplining the Poor: Neoliberal Paternalism and 28 A childcare exemption is reserved for partici- the Persistent Power of Race.” pants who are single custodial parents “caring for a Linda Houser is a doctoral candidate in the Graduate child under six years of age who can prove they are School of Social Work and Social Research at Bryn unable to obtain needed child care within a reason- DECIDING TO DISCIPLINE 417 violence, medical incapacity, outpatient mental er. As suggested by interview and narrative health or substance abuse treatment, and “cir- response data from our current project, case- cumstances beyond [the participant’s] control” workers interpret the requirements and respon- (AWI FG 03-037, 5). In each of these instances, sibilities of this relationship in a variety of ways. with the exception of current or past domestic Moreover, what constitutes pre-penalty coun- violence, the individual must “prove to the RWB seling may include a range of interventions, provider” that she has indeed been rendered from a brief warning or reminder to extensive unable to work. In the case of a physical or referral or direct clinical intervention. In both mental health or substance abuse exemption, the interpretation and application of sanction- such proof is, in principle, limited to docu- ing rules, caseworkers exercise considerable mentation from a licensed physician or recog- discretion. nized mental health or substance abuse Such discretion persists even within the con- professional. However, the ways in which local text of extensive provisions for the training of actors interpret what constitutes adequate proof caseworkers in the application of sanctioning may be mediated by their relationships with rules. Throughout Florida’s 24 regions, case- clients, understandings of their role, and local workers are trained to apply sanctions accord- political and economic contexts. In all instances ing to the Welfare Transition (WT) Sanction for which a good cause exemption is thought to Flow Chart developed by AWI (see Figure A1). be ongoing, participants are required to submit Tracing the various paths that lead to a sanc- to an Alternative Requirement Plan, with the tioning decision suggests some of the ways indi- penalties for failed compliance mirroring those vidual actors might deviate from the planned applied to nonexempt program participants. course. Exactly what these alternative requirements Should an individual be found noncompliant should be is left largely to the discretion of the with the welfare contract, the caseworker is to individual caseworker. mail a Notice of Failure to Participate and Once a participant has been reported for fail- Possible Sanction (form AWI-WTP 2290), com- ure to comply with a work contract require- monly referred to as the “2290,” within two ment, the state requires RWBs to provide notice, days of the first failure. As noted above, an both in writing and orally, of the penalties attempt to contact the participant orally is also attached to noncompliance prior to actually required, and, even if this oral attempt proves imposing such penalties. During this phase of unsuccessful, the participant is allowed 10 days contact, the pre-penalty phase, participants may from the date of the 2290 mailing to establish still avoid a sanction by either establishing good good cause for the noncompliance. A sanction cause for the noncompliance or returning imme- is to be requested only if both attempts at con- diately to satisfactory compliance. Throughout tact fail over a period of 10 days following the this pre-penalty phase, particular emphasis is written notification. If, however, the partici- placed on caseworkers providing counseling, a pant responds to either form of contact within service that includes both reminding or warn- the allotted time period, the pre-penalty phase ing the participant of the consequences of non- can be ended with compliance, provided that one compliance and offering services or supports of two conditions is met: (1) the participant intended to remedy its causes. establishes good cause based on any of the cri- While TANF participants are clearly respon- teria outlined above; or (2) the participant sible for establishing proof of the existence of “agrees to demonstrate satisfactory compliance” grounds for a good cause exemption, program (AWI FG 03-037, 7) and remains compliant guidelines emphasize an ongoing relationship with her welfare contract for a minimum of 30 between the participant and the RWB repre- days following the first failure. sentative, most often the individual casework- Immediate, full-family sanctions are consid- ered appropriate when a participant has either (1) failed to respond to the 2290 and oral able distance from their home or worksite, child care attempts at contact, (2) failed to establish good by a relative or others is unavailable or unsuitable, or cause and refused to demonstrate satisfactory there is no affordable formal child care” (AWI FG 03- compliance, or (3) failed to follow her welfare 037, 7). contract without good cause for a second time 418 AMERICAN SOCIOLOGICAL REVIEW

Figure A1.—Florida Sanction Flow Chart Source: Florida Agency for Workforce Innovation (http://www.floridajobs.org/PDG/TrainingPresentations/WT_SancFlowChart061406_070306.ppt). within 30 days of her first offense. All pre- a guide to when discretion may be most easily penalty and sanctioning activities are to be and least riskily applied. recorded, in accordance with the WT flow chart, While the flow chart provides a basis for sug- in the computerized One Stop Service Tracking gesting that sanctioning in Florida is a highly (OSST) system. With its system-generated structured process, a close reading of the chart prompts and reminders of what actions are shows that substantial opportunities for case- called for and when, the OSST system functions, worker discretion remain. How that discretion if not as a check on discretion, then certainly as gets used is the focus of our analysis. DECIDING TO DISCIPLINE 419

Table A1.—Variable Definitions and Data Sources for Analyses Presented in Tables 4 and 5

Minimum- Independent Variables Definition Maximum Individual Characteristics —Age of client Client age (in years) 18–72 —Number of children Number of children in TANF family 0–12 —Age of youngest child Age of youngest child in TANF family 0–17 —Income Earned income in 1,000s 0–200 —Education (reference category = more than 12 years): —Less than high school 1 = more than 12 years, 0 = otherwise 0–1 —High school 1 = 12 years, 0 = otherwise 0–1 —Black client 1 = black, 0 = otherwise 0–1 —Hispanic client 1 = Hispanic, 0 = otherwise 0–1

Political Environment —County conservatism index Measure of county political ideology, based on factor analysis of –2.5–2.2 local election results for 18 ideologically-relevant constitutional amendments (see Fording et al. 2007). —Percent black Percentage of blacks in county of client in 2000 (County and City 2.1–57.1 Data Books 2003). —Percent Hispanic Percentage of Hispanics in county of client in 2000 (County and 1.5–57.3 City Data Books 2003).

Socioeconomic Environment —Annual wage Average annual income in 1997 for employees in NAICS 7.795–16.674 subsector 722, in 1,000s (County and City Data Books 2003). —Unemployment rate Unemployment rate in county of client, measured each month 1.7–19.7 (Florida Research and Economic Database). —Poverty rate County poverty rate for all persons in 2000 (U.S. Census Bureau 6.9–24.2 Small Area Income and Poverty Estimates). —TANF caseload Number of TANF recipients per 100,000 county residents .142–6.907 (calculated by authors). —Population Total county population in 2000, in millions (County and City .007–2.253 Data Books 2003). Source: Data on client characteristics provided by the Florida Department of Children and Families. Note: TANF = Temporary Assistance for Needy Families; NAICS = North American Industry Classification System. Descriptive statistics are provided for the combined sample and include data for white, black, and Latino welfare clients. 420 AMERICAN SOCIOLOGICAL REVIEW

Table A2.—Regression Results for First-Stage Selection Equation

Independent Variables ␤ Individual Characteristics —Age of client –.0212** —Age of youngest child –.0212 —Black client .3411** —Citizenship status (1 = citizen) .2052** —Education .0358** —Income (in 1000s) –.0394** —Number of children .0387** —Prior sanction (1st spell) .2632**

Community Characteristics —Annual wage – food service/drinking places –.0159 —Local conservatism –.0466** —Percent black –.0038** —Poverty rate –.0399** —TANF caseload t – 1 .2085** Sample size (first spell clients) 40,891 Number of clients returning for 2nd spell 18.827 Note: The sample for this analysis consists of all new TANF clients (single-parent, female, black or white) who entered TANF from January 2001 through December 2002. The dependent variable is a dichotomous variable equal to 1 for clients who returned to TANF for a second spell during our observation period (and 0 otherwise). Cell entries are coefficients generated from an exponential discrete choice model (see Boehmke et al. 2006), estimated using the DURSEL procedure in Stata 10.0 (Boehmke 2005). * p < .05; ** p < .01.

REFERENCES Boehmke, Frederick J., Daniel S. Morey, and Megan Shannon. 2006. “Selection Bias and Continuous- Agency for Workforce Innovation (AWI). 2004, Time Duration Models: Consequences and a January 21. “Final Guidance: Welfare Transition Proposed Solution.” American Journal of Political Program Work Penalties and Pre-Penalty Science 50:192–207. Counseling” (AWI FG 03-037). Alport, Gordon. 1954. The Nature of Prejudice. Bonds, Michael. 2006. “The Continuing Significance Cambridge, MA: Addison-Wesley. of Race: A Case Study of the Impact of Welfare Ayres, Ian and Peter Siegelman. 1995. “Race and Reform.” Journal of African American Studies Gender Discrimination in Bargaining for a New 9:18–31. Car.” American Economic Review 85:304–21. Born, Catherine, Pamela Caudill, and Melinda Bertrand, Marianne and Sendhil Mullainathan. 2003. Cordero. 1999. “Life After Welfare: A Look at “Are Emily and Greg More Employable than Sanctioned Families.” Baltimore, MD: University Lakisha and Jamal? A Field Experiment on Labor of Maryland, School of Social Work. Market Discrimination.” Poverty Action Lab Paper Botsko, Christopher, Kathleen Snyder, and Jacob #3. Retrieved July 20, 2007 (http://povertyaction- Leos-Urbel. 2001. “Recent Changes in Florida lab.org/papers/bertrand_mullainathan.pdf). Welfare and Work, Child Care, and Child Welfare Blalock, Hubert M. 1967. Toward a Theory of Systems.” New Federalism: National Survey of Minority Group Relations. New York: Wiley. America’s Families, State Update no. 2. Bloom, Dan and Donald Winstead. 2002. Sanctions Washington, DC: Urban Institute. and Welfare Reform. Welfare Reform and Beyond Cho, Chung-Lae, Christine A. Kelleher, Deil S. Initiative Policy Brief #12. Washington, DC: Wright, and Susan Webb Yackee. 2005. Brookings Institution. Retrieved October 10, 2007 (http://www.mdrc.org/publications/191/policy- “Translating National Policy Objectives into Local brief.html). Achievements across Planes of Governance and Boehmke, Frederick J. 2005. DURSEL: A Program among Multiple Actors: Second-Order Devolution for Duration Models with Sample Selection (Stata and Welfare Reform Implementation.” Journal of version). Version 2.0. Iowa City, IA: University of Public Administration Research and Theory Iowa. Retrieved September 3, 2007 15:31–54. (http://myweb.uiowa.edu/fboehmke/methods). Correll, Sherry, Stephen Benard, and In Paik. 2007. DECIDING TO DISCIPLINE 421

“Getting a Job: Is There a Motherhood Penalty?” Research Paradigm.” Perspectives on Politics American Journal of Sociology 112:1297–1338. 5:63–79. Crenshaw, Kimberle. 1991. “Mapping the Margins: Hasenfeld, Yeheskel, Toorjo Ghose, and Kadyce Intersectionality, Identity Politics, and Violence Larson. 2004. “The Logic of Sanctioning Welfare Against Women of Color.” Stanford Law Review Recipients: An Empirical Assessment.” Social 43:1241–99. Service Review 78:304–19. Darley, John M. and Paget H. Gross. 1983. “A Kalil, Ariel, Kristin S. Seefeldt, and Hui-chen Wang. Hypothesis-Confirming Bias in Labeling Effects.” 2002. “Sanctions and Material Hardship under Journal of Personality and Social Psychology TANF.” Social Service Review 76:643–62. 44:20–33. Keech, William R. [1968] 1981. The Impact of Negro Devine, Patricia G. and Sarah M. Baker. 1991. Voting: The Role of the Vote in the Quest for “Measurement of Racial Stereotype Subtyping.” Equality. Chicago, IL and Westport, CT: Rand Personality and Social Psychology Bulletin McNally and Greenwood. 17:44–50. Keiser, Lael R., Peter Meuser, and Seung-Whan Dias, Janice Johnson and Steven Maynard-Moody. Choi. 2004. “Race, Bureaucratic Discretion, and 2007. “For-Profit Welfare: Contracts, Conflicts, the Implementation of Welfare Reform.”American and the Performance Paradox.” Journal of Public Journal of Political Science 48:314–27. Administration Research and Theory 17:189–211. Key, V. O., Jr. 1949. Southern Politics in State and Duster, Troy. 2008. “Introduction to Unconscious Nation. New York: Knopf. Racism Debate.” Social Psychology Quarterly Kinder, Donald R. and Thomas Palfrey, eds. 1993. 71(1):6. Experimental Foundations of Political Science. Eberhardt, Jennifer L., Paul G. Davies, Valerie J. Ann Arbor, MI: University of Michigan Press. Purdie-Vaughns, and Sheri Lynn Johnson. 2006. Koralek, Robin. 2000. “South Carolina Family “Looking Deathworthy: Perceived Stereotypicality Independence Program Process Evaluation.” of Black Defendants Predicts Capital Sentencing Prepared for South Carolina Department of Social Outcomes.” Psychological Science 17:383–86. Services. Washington, DC: The Urban Institute. Fording, Richard C., Joe Soss, and Sanford F. Schram. Lieberman, Robert C. 1998. Shifting the Color Line: 2007. “Devolution, Discretion, and the Effect of Race and the American Welfare State. Cambridge, Local Political Values on TANF Sanctioning.” MA: Harvard University Press. Social Service Review 81:285–316. Lindhorst, Taryn, Ronald J. Mancoske, and Alice A. Fox, Cybelle. 2004. “The Changing Color of Welfare? Kemp. 2000. “Is Welfare Reform Working? A How Whites’Attitudes towards Latinos Influence Study of the Effects of Sanctions on Families Support for Welfare.” American Journal of Receiving TANF.” Journal of Sociology and Social Sociology 110:580–625. Welfare 27(4):185–201. Gilens, Martin. 1999. Why Americans Hate Welfare: Lipsky, Michael. 1980. Street-Level Bureaucracy: Race, Media, and the Politics of Antipoverty Policy. Dilemmas of the Individual in Public Services. Chicago, IL: University of Chicago Press. New York: Russell Sage. Goldberg, Heidi and Liz Schott. 2000. “A Loprest, Pamela J. 2002. “Making the Transition Compliance Oriented Approach to Sanctions in from Welfare to Work: Successes by Continuing State and County TANF Programs.” Washington, Concerns.” Pp. 17–31 in Welfare Reform: The Next DC: Center on Budget and Policy Priorities. Act, edited by A. Weil and K. Finegold. Goldrick-Rab, Sara and Kathleen M. Shaw. 2005. Washington, DC: The Urban Institute Press. “Racial and Ethnic Differences in the Impact of Mancuso, David and Vanessa L. Lindler. 2001. Work-First Reforms on Access to Postsecondary “Examining the Circumstances of Welfare Leavers Education.” Educational Evaluation and Policy and Sanctioned Families in Sonoma County Final Analysis 27:291–307. Report.” SPHERE Institute, Burlingame, CA. Gooden, Susan Tinsley. 2003. “Contemporary Maynard-Moody, Steven and Michael Musheno. Approaches to Enduring Challenges: Using 2003. Cops, Teachers, Counselors: Stories from the Performance Measurement to Promote Racial Front Lines of Public Service. Ann Arbor, MI: Equality under TANF.” Pp. 254–78 in Race and the University of Michigan Press. Politics of Welfare Reform, edited by S. F. Schram, Mead, Lawrence M., ed. 1997. The New Paternalism: R. C. Fording, and J. Soss. Ann Arbor, MI: Supervisory Approaches to Poverty. Washington, University of Michigan Press. DC: Brookings Institution Press. Hancock, Ange-Marie. 2004. The Politics of Disgust: Munnell, Alicia H., Geoffrey M. B. Tootell, Lynn E. The Public Identity of the Welfare Queen. New Browne, and James McEneaney. 1996. “Mortgage York: New York University Press. Lending in Boston: Interpreting HMDA Data.” ———. 2007. “When Multiplication Doesn’t Equal American Economic Association 86:25–53. Quick Addition: Examining Intersectionality as a Neubeck, Kenneth and Noel Cazenave. 2001. Welfare 422 AMERICAN SOCIOLOGICAL REVIEW

Racism: Playing the Race Card Against America’s Burke, Sankey Williams, John M. Eisenberg, Josee Poor. New York: Routledge. J. Escarce, and William Ayers. 1999. “The Effect Pager, Devah. 2007. Marked: Race, Crime, and of Race and Sex on Physicians’Recommendations Finding Work in an Era of Mass Incarceration. for Cardiac Catheterization.” New England Journal Chicago, IL: University of Chicago Press. of Medicine 340:618–26. Pavetti, LaDonna A., Michelle K. Derr, and Heather Shaw, Kathleen M., Sara Goldrick-Rab, Christopher Hesketh. 2003. “Review of Sanction Policies and Mazzeo, and Jerry A. Jacobs. 2006. Putting Poor Research Studies: Final Literature Review.” Report People to Work: How the Work-First Idea Eroded prepared for ASPE. Washington, DC: Mathematica College Access for the Poor. New York: Russell Policy Research. Sage Foundation. Pavetti, LaDonna A., Michelle K. Derr, Gretchen Soss, Joe, Richard C. Fording, and Sanford F. Schram. Kirby, Robert G. Wood, and Melissa A. Clark. 2008. “The Color of Devolution: The Politics of 2004. “The Use of TANF Work-Oriented Sanctions Local Punishment in an Era of Neo-liberal Welfare in Illinois, New Jersey, and South Carolina: Final Reform.” American Journal of Political Science Report.” Report prepared for ASPE. Washington, 52:536–53. DC: Mathematica Policy Research. Soss, Joe, Sanford F. Schram, Thomas Vartanian, Quadagno, Jill. 1994. The Color of Welfare: How and Erin O’Brien. 2001. “Setting the Terms of Racism Undermined the War on Poverty. New Relief: Explaining State Policy Choices in the York: Oxford University Press. Devolution Revolution.” American Journal of Quillian, Lincoln. 2008. “Does Unconscious Racism Political Science 45:378–95. Exist.” Social Psychology Quarterly 71(1):6–11. Stahl, Roland. 2008. “Examining TANF Sanction Rector, Robert E. and Sarah E. Youssef. 1999. “The Policies: Who Gets Sanctioned and What Are the Determinants of Welfare Caseload Decline.” Effects?” Phd Dissertation, Graduate School of Report no. 99–04. Washington, DC: Heritage Social Work and Social Research, Bryn Mawr Foundation. College, Bryn Mawr, PA. Reese, Ellen. 2005. Backlash against Welfare Starobin, Paul. 1998. “The Daddy State.” National Mothers: Then and Now. Berkeley, CA: University Journal 28(March):678–83. of California Press. Valentino, Nicholas A., Vincent Hutchings, and Ismail Reichman, Nancy E., Julian O. Teitler, and Marah A. K. White. 2002. “Cues that Matter: How Political Curtis. 2005. “TANF Sanctioning and Hardship.” Ads Prime Racial Attitudes During Campaigns.” Social Service Review 79(2):215–36. American Political Science Review 96:75–90. Richards, Zoe and Miles Hewstone. 2001. “Subtyping and Subgrouping: Processes for the Prevention Wacquant, Loic. 2001. “The Penalisation of Poverty and Promotion of Stereotype Change.” Personality and the Rise of Neo-Liberalism.” European and Social Psychology Review 5:52–73. Journal on Criminal Policy and Research Schneider, Anne and Helen Ingram. 1997. “The 9:401–11. Social Construction of Target Populations: Watkins-Hayes, Celeste. 2009. The New Welfare Implications for Politics and Policy.” American Bureaucrats: Situated Bureaucrats and Political Science Review 87:334–47. Entanglements of Race, Class, and Welfare. Schram, Sanford F. 2005. “Contextualizing Racial Chicago, IL: University of Chicago Press. Disparities in American Welfare Reform: Toward Weissert, Carol S., ed. 2000. Learning from Leaders: a New Poverty Research.” Perspectives on Politics Welfare Reform Politics and Policy in Five 3:253–68. Midwestern States. Albany, NY: Rockefeller ———. 2006. Welfare Discipline: Discourse, Institute Press. Governance, and Globalization. Philadelphia, PA: Western, Bruce. 2006. Punishment and Inequality in Temple University Press. America. New York: Russell Sage Foundation. Schram, Sanford F., Richard C. Fording, and Joe Wissoker, Douglas, Wendy Zimmermann, and Soss. 2008. “Neoliberal Poverty Governance: Race, George Galster, eds. 1998. Testing for Place and the Punitive Turn in U.S. Welfare Policy.” Discrimination in Home Insurance. Washington, Cambridge Journal of Regions, Economy and DC: Urban Institute Press. Society 1:1–20. Wu, Chi-Fang, Maria Cancian, Daniel R. Meyer, and Schram, Sanford F., Joe Soss, and Richard C. Fording. Geoffrey Wallace. 2006. “How Do Welfare 2003. Race and the Politics of Welfare Reform. Sanctions Work?” Social Work Research 30:33–51. Ann Arbor, MI: University of Michigan Press. Yinger, John. 1995. Closed Doors, Opportunities Schulman, K. A., Jesse A. Berlin, William Harless, Lost: The Continuing Costs of Housing Jon F. Kerner, Shyrl Sistrunk, Bernard J. Gersh, Discrimination. New York: Russell Sage Ross Dubee, Christopher K. Taleghani, Jennifer E. Foundation.